Supplementary information Methods 1. Creation of the plastid dataset We retrieved the protein annotations for 75 selected plastid genomes of Rhodophyta, Cryptophyta, Haptophyceae and Ochrophyta from the NCBI RefSeq database (https://www.ncbi.nlm.nih.gov/) (Supplementary Table 8). We used OrthoFinder (Emms and Kelly 2015) with a BLASTP E-value threshold of 1e-5 and an MCL inflation parameter of 1.5 to produce orthogroups (OGs). We filtered the 504 resulting OGs to retain those (108) containing ≥20 species (of which ≥1 Rhodophyta, ≥1 Stramenopiles, and either ≥1 Cryptophyta or ≥1 Haptophyceae). We first aligned the selected OGs with MAFFT (L-INS-i algorithm, 5000 iterations) (Katoh and Standley 2013), then enriched them by adding more species from genomic data (such as the five new species sequenced in this study) with Forty-Two (https://metacpan.org/release/Bio-MUST-Apps-FortyTwo). We checked for possible paralogy using methods that are described in the section about the construction of the nuclear dataset (see below) and found only one dubious OG, from which we manually removed four paralogous sequences. We further discarded 9 additional OGs with <30 species. Finally, to select unambiguously aligned positions, we applied a loose BMGE (Criscuolo et al. 2010) filter (entropy cutoff of 0.6 and gap cutoff of 0.4) on each aligned OG. 2. Creation of the mitochondrial dataset As for the plastid, we retrieved all the protein annotations available for stramenopiles mitochondrial genomes from the NCBI website (Supplementary Table 9). To this first set, we added the annotations of the five new species generated in this study, as well as some identified from genomic scaffolds of Labyrinthulomycetes and Xanthophyceae species presenting a high similarity to mitochondrial genomes, using MFannot server (Beck and Lang 2010; MFannot, organelle genome annotation websever; http://megasun.bch.umontreal.ca/cgi-bin/mfannot/mfannotInterface.pl). We chose to integrate all annotations before delineating OGs because it can be more difficult to retrieve orthologs for fast evolving mitochondrial sequences with Forty-Two. We generated OGs using the same protocol as with the plastid dataset and retained the 33 OGs containing ≥66% of the species of the dataset. We aligned the clusters with MAFFT (L-INS-i algorithm, 5000 iterations) and manually fixed frameshift errors for some sequences of Fragilariopsis and Aurantiochytrium in the resulting alignments. We also split the sequence of Phaeodactylum fusion protein nad9-rps14 and added each half to its respective alignment. Finally, we removed one OG showing a too ambiguous alignment and applied BMGE on each aligned OG as above. 3. Generation and selection of orthogroups for the nuclear dataset We retrieved the complete proteomes for 53 species across Stramenopiles, Alveolata and Rhizaria from the NCBI website (Supplementary Table 10). We performed pairwise similarity 33 searches between all proteomes using USEARCH (Edgar 2010) with a minimal value of 1e-5 (instead of BLAST to speed up computations). We then generated OGs with OrthoFinder, as for the plastid and mitochondrial datasets. Out of a total of 212,849 OGs, we retained only those containing ≥15 species, with at least one species from each of the three clades (Stramenopiles, Alveolata and Rhizaria) and ≤600 sequences to avoid retrieving large protein families. These filters left us with 3,063 OGs. At this stage, we observed that some OGs contained highly divergent sequences (possibly non-homologous), dragged into the clusters by a single similarity link. To address this issue, we removed the sequences that were not similar to a minimum percentage of the other sequences in each OG (BLAST E-value threshold of 1e-10). We proceeded in two steps, first removing the sequences matching <30% of the other sequences, then 50%. These two steps removed 17,734 and 6,335 sequences, respectively. Finally, we filtered the OGs anew, so as to retain only those with ≥15 species, hence reducing their number to 2,892. To classify OGs among those close to true orthogroups and those corresponding to more complex multigene families, we used an automated phylogenetic analysis of single-gene trees (Simion et al., 2017). Briefly, we first aligned OGs with MAFFT v7 (L-INS-I algorithm, 5000 iterations) (Katoh and Stanley 2013), then filtered out columns with <5% of amino acid residues and sequences with <50 parsimony informative positions. We then inferred trees with RAxML v8 (Stamatakis 2014) using the LG+F+Γ4 model (Le and Gascuel 2008), and ran a custom script aimed at detecting cases of ancient paralogy. As in Simion et al. (2017), we computed how many of seven predefined clades (Rhizaria, Ciliophora, Myzozoa, Oomycetes, Labyrinthulomycetes, Blastocystis and Ochrophyta) were affected by out-paralogy (i.e., at least two clans containing sequences exclusively from a given clade). This allowed us to separate the OGs containing ≤3 out-paralogs (1904) from the other, more complex OGs (988) having more out-paralogs. To split the OGs showing too many cases of out-paralogy (which may correspond to an ancient gene duplication), we used the software root-max-div (Simion et al. 2017). This program searches for the branch maximizing the taxonomic diversity on both sides of the bipartition and splits the tree on the branch if (i) the number of sequences on each side satisfies a minimal threshold (two first parameters), (ii) the number of common species on both sides is above a minimal threshold (third parameter) and (iii) the branch length is among the top percentile of the tree branches (fourth parameter). We applied four different parameter sets in the following order (30-30-0-5, 30-10-0-5, 10-10-0-5, and 40-10-0-20), retrieving the two sub-alignment files of the first successful parameter set for each OG. We repeated the whole procedure until no more gene tree could be cut. Finally, we filtered the split OGs again, to retain only those with ≥15 species, thereby reducing their number to 336. Finally, we pooled the two groups of OGs, yielding a total of 2,240 OGs, of which we reduced the redundancy with a custom script targeting highly similar subsequences of the same organism inside individual OGs, as in Simion et al. (2017). This step removed 5,573 sequences. 34 4. Assessment of transcriptome quality Before improving the taxon sampling of our nucleus dataset using a combination of transcriptomic and genomic data, we evaluated the contamination level of the available transcriptomes. They consisted in assemblies from MMETSP (Keeling 2014), TSA retrieved from the NCBI website, and SRA raw reads also retrieved from the NCBI website, which were assembled using Trinity v2.6 (Haas 2013) with the trimmomatic and jaccard clip options (--jaccard_clip --trimmomatic). A set of 80 highly expressed gene alignments (ribosomal proteins), on which a large diversity of eukaryotic sequences are regularly added and manually curated, was used as a reference. To estimate the contamination level, we took advantage of Forty-Two and its taxonomic filters to add back previously incorporated organisms to this dataset. Forty-Two was designed to search transcriptomes for orthologous sequences and add them to existing alignments. At this last step, it can verify if the added orthologous sequence satisfies a user-defined positive and/or negative taxonomic filter (i.e., belonging to Stramenopiles or not belonging to Xanthophyceae). Here, we checked that added sequences indeed matched an organism of the same genus in the alignment, which allowed us to distinguish between orthologous sequences genuinely belonging to the transcriptome from orthologous sequences belonging to contaminants. We considered a transcriptome to be clean when we found <5 contaminant sequences over the 80 alignments. For each contaminant sequence, we further retrieved the most closely related organism in the alignment, so as to design optimal taxonomic filters for contaminated transcriptomes. Overall, this approach allowed us to exploit taxonomically interesting transcriptomes that were contaminated without adding contaminated sequences in our OGs. 5. Vaucheria litorea transcriptome decontamination Whereas Vaucheria litorea was one of the only Xanthophyceae for which a large amount of data was available, we observed that its transcriptome was contaminated by a large array of organisms. Because of its isolated phylogenetic position in our current sample of the eukaryotic diversity, combined to a relatively fast evolutionary rate, it was difficult to only rely on sequence similarity for decontamination. Thus, to tackle this issue, we implemented a strategy based on k-mer distributions (Teeling et al. 2004) to identify and remove the largest part of Vaucheria contaminant sequences. Briefly, we assembled two sets of Vaucheria sequences (i.e., genuine and contaminant), for which we computed the frequencies of all possible 6-nt k-mers. Then, the k-mer composition of each transcript was compared against those reference distributions using an Euclidean distance, and we discarded the transcripts closer to the contaminant than the genuine sequences. To define these two sets of reference, we used eukaryotic orthologs (594 nuclear genes from 370 species covering the diversity of eukaryotes) from a non-published study, in which we added Vaucheria transcripts with Forty-Two. (We used these orthologs instead of the 80 ribosomal proteins to maximize the number and the variety of sequences in the reference sets.) Then, we inferred the single-gene phylogeny of all orthologs using RAxML (LG+F+Γ4 model) and retrieved the taxonomy of the sister clan of each Vaucheria sequence. We considered transcripts added close to Ochrophyta species as genuine reference sequences, whereas the other transcripts were pooled separately as different sources of contaminants. Those were mainly 35 representatives of Labyrinthulomycetes, Discosea and Viridiplantae. To identify contaminants in the full Vaucheria transcriptome, we first tested the transcripts against Labyrinthulomycetes contaminated sequences then to Discosea sequences, and finally to Viridiplantae sequences.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages27 Page
-
File Size-